Cluster randomized trials are trials in which intact units such as hospitals, clinics, or schools, are randomly assigned to a condition such as Treatment or Control. Cluster randomized trials play a prominent role in medicine, health, and social policy research, accounting for some 11,000 trials in these areas, and the use of these trials has been accelerating in recent years. The statistical procedures required to plan and to analyze a cluster randomized trial are substantially more complex than those used for simple randomized trials. Popular software packages such as SPSS, SAS, and Stata include modules for the analysis of cluster randomized trials, and some 80% of these trials are analyzed using these kinds of modules. By contrast, no commercial software exists to compute power for these trials. Therefore, only about 20% of these trials use the proper techniques to compute power when these studies are being planned. This is a serious problem, since power for the cluster randomized design typically drops by at least 50%, and often by more, as compared with a simple randomized design with the same total sample size. The goal of this project is to develop software to compute power for cluster randomized trials. The program will work with studies that include two, three or four levels in the hierarchy. It will incorporate a sophisticated user interface to address the kinds of real-world issues that must be addressed in a power analysis. On the input side, it will allow users to enter the kinds of data that they are likely to have, and to do so using a clear and intuitive interface. On the output side the program will create tables and graphs that allow the user to quickly assess the impact of all assumptions, and to determine how changes in the design would affect power and costs.
Cluster randomized trials are trials in which intact units such as hospitals, clinics, or schools, are randomly assigned to a condition such as Treatment or Control. These kinds of trials play a prominent role in medicine, health, and social policy research, with as many as 1,000 such trials being planned each year. The goal of this project is to develop software to perform a power analysis for cluster randomized trials, and enable researchers to design these trials in a manner that yields maximum power for the smallest cost.
|Borenstein, Michael; Higgins, Julian P T (2013) Meta-analysis and subgroups. Prev Sci 14:134-43|